Introduction

uuldana-cli is an open-source, provider-agnostic supervisor for native AI coding agents. It lets you work with the agents you already use through one terminal workflow, in the order you choose, and keeps the work going when one of them reaches its usage quota.

The agent still does the coding. uuldana-cli starts the agent’s own terminal interface, relays your keyboard to it unchanged and draws its screen under a fixed header, records what happens in a journal, and holds the permission ceiling you set. It does not supply a model or replace an agent’s reasoning loop.

Native behavior comes first

When an agent is first in your list, you are using the real agent, in its own terminal interface. Agents keep their own tools, instruction files, authentication, plugins, skills, and MCP configuration. uuldana-cli does not read credentials and does not write provider configuration.

What works today

This is an early version (v0.1.0 on npm). The command line does the following:

• uuldana starts the first available agent in the current directory, in its native interface, under a header that always shows uuldana-cli and which agent is working. It needs no configuration: it finds the agents installed on your machine by itself.
• /agent, typed on an empty line, lets you choose which agent works first and which ones take over.
• uuldana resume continues the last session of the current directory from its journal.
• uuldana providers lists the agents found on this machine and what was learned about each.
• When the active agent prints its usage-limit message, uuldana-cli stops it gracefully and continues the work on the next one, without Ctrl+C and without retyping.
• uuldana --bypass-permissions starts a run without approval prompts or limits, for that run only.

Four agents are installed on the test machine. uuldana-cli found all four by itself, with no configuration, and ran each through itself against its real installation. Agents that are not installed there have not been tried. Current implementation status lists exactly what was tested and what was not.

Where to go next

• Core concepts defines providers, transports, sessions, and fidelity.
• Installation and Initial configuration get you to a first run.
• Automatic failover explains what happens at a usage limit.
• Plugins, skills & MCP describes what carries over when work moves between providers, and what does not.